V. Balasubramanian

This paper shows that learning in physical systems leaves imprints on the Hessian, revealing task-specific changes in system response and susceptibility.

The paper shows that simpler models of the world lead to better fitness in bacterial populations under limited information.

The paper investigates how CRISPR-Cas systems in prokaryotes may cause autoimmunity through cross-reactive spacer targeting, revealing a scaling relationship between spacer length and CRISPR repertoire size.

The paper examines the trade-off between the size of the bacterial immune repertoire and the effectiveness of response to phage threats, showing how this limits the optimal size of CRISPR cassettes.

The paper models the adaptive immune system as a Bayesian framework that balances new pathogen observations with prior experience to predict future threats and optimize memory repertoire updates.

The paper develops a model of two-stage learning in neural circuits, showing how mismatches between tutor signals and plasticity mechanisms can impair learning, and applies it to birdsong.